Mannil M et al, 2018: Three dimensional texture analysis with machine learning provides incremental predictive information for successful shock wave lithotripsy in patients with kidney stones.
Mannil M, von Spiczak J, Hermanns T, Poyet C, Alkadhi H, Fankhauser CD.
Institute of Diagnostic and Interventional Radiology and Department of Urology (TH, CP, CDF), University Hospital Zurich, University of Zurich, Switzerland.
Abstract
PURPOSE: To determine the predictive value of three-dimensional texture analysis (3D-TA) in computed tomography (CT) images for successful shock wave lithotripsy (SWL) in patients with kidney stones. MATERIAL AND METHODS: Patients with pre and postoperative CT scans, previously untreated kidney stones and a stone diameter of 5-20 mm were included. A total of 224 3D-TA features of each kidney stone, including the attenuation measured in Hounsfield Units (HU), and the clinical variables body mass index (BMI), initial stone size, and skin-to-stone distance (SSD) were analyzed using five commonly used machine learning models. The data set was split in a ratio of 2/3 for model derivation and 1/3 for validation. Machine learning-based predictions for SWL success in the validation cohort were evaluated calculating sensitivity, specificity, and the area-under-the-curve (AUC).
RESULTS: For SWL success the three clinical variables BMI, initial stone size and SSD showed AUCs of 0.68, 0.58 and 0.63 respectively, but no predictive value for HU was found. A RandomForest classifier using three 3D-TA features had an AUC of 0.79. By combining these three 3D-TA features with clinical variables, the discriminatory accuracy improved further with an AUC of 0.85 for 3D-TA features and SSD, an AUC of 0.8 for 3D-TA features and BMI and an AUC of 0.81 for 3D-TA and stone size.
CONCLUSION: This preliminary study indicates that the clinical variables BMI, initial stone size and SSD show limited value for predicting SWL success, while the HU values of stones were not predictive. Selected 3D-TA features identified by machine learning provided incremental accuracy for predicting the success to SWL.
J Urol. 2018 Apr 16. pii: S0022-5347(18)42986-2. doi: 10.1016/j.juro.2018.04.059. [Epub ahead of print]
Comments 1
In view of the advantage of predicting the outcome of SWL, this report is of great interest.
With an advanced 3D-texture analysis combined with state-of-the-art artificial learning, the authors for the first time showed how this method might be used for prediction of stone disintegration.
The basic conclusion was that stone density in terms of HU is not predictive for SWL-success, whereas limited use was noted for BMI, stone size and SSD (skin-to-stone distance).
Although the 3D-TA method can increase the predictive accuracy of SWL success, the procedure is advanced and requires both special expertise and equipment. Hopefully the method can be of clinical value in the future, but today it is mainly of theoretical interest as a development project, at least in the eyes of the reviewer who not fully understands the details of this technology.